Extended Semi-Supervised Learning GAN for Hyperspectral Imagery Classification
Andrew Hahn, Murali Tummala, James W. Scrofani · 2019
Hyperspectral imagery (HSI) cubes are high-dimensional datasets that lend themselves well to deep learning approaches for classification. Deep learning approaches, specifically generative adversarial networks (GANs), have been shown to be very effective in classification and generation of accurate synthetic data in computer vision problems. This work proposes an extension of an existing GAN training scheme, called extended semi-supervised learning (ESSL), metrics for evaluating GAN training performance, and demonstrates the effectiveness of the proposed training scheme to improve classification of HSI. Using ESSL with GAN, we have been able to achieve approximately 0.8% increase in classification accuracy over convolutional neural networks as well as generate extremely accurate synthetic imagery.